Research On Multivariate Quality Control And Diagnosis Based On Artificial Neural Network

With the development of global economy, the market competition is becomingmore and more intense. Improving the product qualityand keeping the product qualityconformance is one of the most important methods to win in the competition, and themonitoring and optimizing for the production process is an important way to insurethe product quality. Especially in the modern production process of complicatedproducts, there are many quality attributes influencing the product quality together. Sothe combination control of these quality attributes becomes an important aspect tokeep the product quality.Multivariate quality control and diagnosis is about multivariate normal process,and the variables always are not independent but correlated. So if the vector of one orseveral variables, and /or variance, and /or the relationship between the variables doesnot accord with the population, the process is out of control and the control chartsmonitoring the process will show warring signal. Detecting whether the process is outof control or not, especially which variable(s) result(s) in the out-of-control, is just theproblem that the multivariate statistical process control and diagnosis techniques tryto research and solve. This paper is based on several kinds of methods of multivariatequality diagnosis, proposed multivariate quality control and diagnosis based onartificial neural network, and included the following contents:(1) The paper summarizes the general methods of multivariate quality controland diagnosis and the problems. The paper proposes a way to do quality control anddiagnosis, and looking for the out-of-control variable(s) using artificial neuralnetwork in detail.(2) The paper analyses the characteristics of multivariate quality control, andexplains the mean vector control chart theory and the methods and steps ofestablishing theT2 control chart.(3) Multivariate quality control and diagnosis based on artificial neural networkis proposed, because of the characteristics of multivariate quality control anddiagnosis and the advantage of artificial neural network, and the multivariate qualitydiagnosis neural network is designed in detail. First, use the multivariate qualitycontrol chart to detect the abnormal signal. Then, conduct quality diagnosis usingrelated neural network to look for which variable(s) result(s) in the processout-of-control.(4) The paper designs the BP neural network for diagnosis with an example, trains the net and tests the net using the neural network tool of the Matlab software.The research result shows that the method, this paper proposes, is an effectivewayto multivariate qualitycontrol and diagnosis, and it is meaningful for multivariatequalitydiagnosis theory and practice.